Medical Question Answering on MedQA (Acc, ECE, Brier)
66.8AccuracyGRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GRPOBackbone Model=Ministral 3 8B2026.06 | 66.8 | 30.37 | 62.66 | |
| SFTBackbone Model=Ministral 3 8B2026.06 | 65.2 | 33.06 | 67.83 | |
| BaseBackbone Model=Ministral 3 8B2026.06 | 65 | 31.56 | 65.73 | |
| CoRABackbone Model=Ministral 3 8B2026.06 | 64.6 | 31.91 | 65.41 | |
| CoRABackbone Model=Gemma 2 9B Instruct2026.06 | 55.6 | 42.22 | 86.48 | |
| GRPOBackbone Model=Gemma 2 9B Instruct2026.06 | 54.6 | 43.28 | 87.56 | |
| GRPOBackbone Model=Qwen2.5 7B2026.06 | 54.4 | 35.17 | 76.74 | |
| CoRABackbone Model=Qwen2.5 7B2026.06 | 52 | 39.72 | 84.79 | |
| SFTBackbone Model=Gemma 2 9B Instruct2026.06 | 52 | 47.58 | 94.83 | |
| BaseBackbone Model=Gemma 2 9B Instruct2026.06 | 51.8 | 46.86 | 93.86 | |
| SFTBackbone Model=Qwen2.5 7B2026.06 | 50.4 | 44.45 | 91.39 | |
| BaseBackbone Model=Qwen2.5 7B2026.06 | 46.6 | 42.94 | 93.13 |